chibz3 / fifa-19-linear-regression

Using linear regression to predict player values of FIFA 19 data

fifa-project

Flatiron Module 2 Linear regression Project

Project Overview

Working with the FIFA 19 video game data, I'll be analyzing, the different aspects of the players game atributes to build a linear regression model that would predicit market value of the players in the top 5 leagues in Europe. The English, German, French, Italian and Spanish leagues.

Approch

• my process had 6 parts:

• Getting my Data

• Cleaning and analyzing the data using Pandas

• Running Statistical tests on my data using scipy

• Visualizing our insights using Seaborn and Mathplotlib

• Engineering Features for my model

• Fiting my model

Getting my data

From Kaggle, I got the fifa 19 game data, complete with all the different attributes for each player.Each row contained each player in the game and the columns contain different information about them.

Cleaning and analyzing the data using Pandas

I spent sevral hours cleaning the data using Pandas Dataframes and Series to prepare it for analysis. I removed or updated Null values. I group relevating information together and created new columns When necessary. I was able to select a data frame with just the players belonging to the top five leagues in Europe.I also changed the datatypes of many columns so I could make calculations and comparisions acorss columns.

Running Statistical tests on my data using Scipy

I ran sevral statistical test on my data using sciypy. For example, I ran an ANOVA test to see if the average age was the same accross the different leagues.

Visualizing our insights using Seaborn and Mathplotlib

I explored my data and provided various visualiztions for the statistical tests I carried out and also to gain more insigt on the factors I considered to have and effect on how much a player is valued.

Engineering Features for my model

I created new features from the information in my data. I was able to uptain knowlege from the EDA on things that affect the value of my data. I checked for coorelation and dropped features that were coorelated to the value which is my target varribale. I transformed non linear relationships with my target varriable to better capture a linear relationship so my model would fit accordingly.

Fiting my model with ols (statsmodels) and sklearn Lasso regressoion

After all the feature selection, I split my data so I could have a test sample after I fit my model to check its acurracy. I carried severally fits with ols from statsmodel, Lasso and the ordinary linear regression model form sklearn. I went on the sepreate my data set to outfield players and Goalkeepers and fit two different models to test for better accuracy.

Conclusion

I was able to fit my model to R square greater that 0.90 and root mean square error within 0.4 of the standard devation of the test sample, which was a very good model.

Moving Forward

FIFA does alot of work to mold a players real world physical and mental attribute into the game. Moving forward my hope is that I can use this insight I got about different Important Atributes FIFA uses to determine Player values to Apply it to real world player values. I believe if one is able to capture cetain attributes and informtion about a player in the real world, in conjuction with real world Market pointers, One can predict actual player Market Value

Presentation

https://github.com/chibz3/fifa-project/blob/master/fifa-project_presentation%20.pdf